Autogen MCP : Error: unhandled errors in a TaskGroup (1 sub-exception) using Autogen and gpt-4o.

[appuk45]

What happened?

Describe the bug
I faced a bug when I tried to use Autogen with the Shopify MCP server. The model calls the tool correctly, including the correct query. It is working fine when tried using langchain and MCP Inspector.

To Reproduce

import asyncio

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import StdioServerParams, mcp_server_tools
from autogen_agentchat.ui import Console
import os
from dotenv import load_dotenv
load_dotenv()

async def main() -> None:
# Get the fetch tool from mcp-server-fetch.
shopify_mcp_server = StdioServerParams(command="node", args=["shopify-mcp-server/build/index.js"],env={'SHOPIFY_ACCESS_TOKEN':#shopify_access_token ,'MYSHOPIFY_DOMAIN':#shopify_domain})
tools = await mcp_server_tools(shopify_mcp_server)

# Create an agent that can use the fetch tool.
model_client = OpenAIChatCompletionClient(model="gpt-4o",api_key=#api_key)
agent = AssistantAgent(name="shopify_assistant", model_client=model_client, tools=tools, reflect_on_tool_use=True)  # type: ignore
# Let the agent fetch the content of a URL and summarize it.
await Console (agent.run_stream(task="show order #1001"))
# print(result.messages[-1].content)

asyncio.run(main())

Current Behavior
When I asked to show an order of order number 100 it is showing this output:
Shopify MCP Server running on stdio
---------- user ----------
show order 100
---------- shopify_assistant ----------
[FunctionCall(id='call_2gEpQguofxe0Z5cLkLYNg3yk', arguments='{"query":"name:100","first":1}', name='get-orders')]
Shopify MCP Server running on stdio
---------- shopify_assistant ----------
[FunctionExecutionResult(content='Error: unhandled errors in a TaskGroup (1 sub-exception)', name='get-orders', call_id='call_2gEpQguofxe0Z5cLkLYNg3yk', is_error=True)]
---------- shopify_assistant ----------
It seems there was an error processing your request to fetch order 100. Could you please verify the order number or check if there are any connectivity issues? Let me know if you would like me to try again or assist further.

When I give the orderId instead or order_number the model is fetching the correct result. But for most of the queries the same issue is happening.

Which packages was the bug in?

Python Extensions (autogen-ext)

AutoGen library version.

Python dev (main branch)

Other library version.

No response

Model used

gpt-4o

Model provider

OpenAI

Other model provider

No response

Python version

None

.NET version

None

Operating system

Windows

[ekzhu]

Without having access to shopify I can't reproduce the result. But have you tried other MCP servers? Do they all of the same error? Your help in debugging this would be great.

[ekzhu]

[FunctionCall(id='call_2gEpQguofxe0Z5cLkLYNg3yk', arguments='{"query":"name:100","first":1}', name='get-orders')]

Does the function call argument look right to you?

[appuk45]

[FunctionCall(id='call_2gEpQguofxe0Z5cLkLYNg3yk', arguments='{"query":"name:100","first":1}', name='get-orders')]

Does the function call argument look right to you?

Yes the function call is correct. But the error is happening when the tool is executing.

[appuk45]

Also

[FunctionCall(id='call_2gEpQguofxe0Z5cLkLYNg3yk', arguments='{"query":"name:100","first":1}', name='get-orders')]

Does the function call argument look right to you?

Also when I ask the model 'show order 12345678' where the orderId is 12345678, it will call the get-order tool and will successfully return the result

[appuk45]

Without having access to shopify I can't reproduce the result. But have you tried other MCP servers? Do they all of the same error? Your help in debugging this would be great.

https://github.com/pashpashpash/shopify-mcp-server

This is the Shopify MCP server. This is working fine in langchain and MCP inspector

[appuk45]

Without having access to shopify I can't reproduce the result. But have you tried other MCP servers? Do they all of the same error? Your help in debugging this would be great.

Tried it with the Github MCP server.

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import StdioServerParams, mcp_server_tools
from autogen_agentchat.ui import Console
import os 
from autogen_core import CancellationToken
from dotenv import load_dotenv
load_dotenv()

async def main() -> None:
    # Get the fetch tool from mcp-server-fetch.
    github_mcp_server = StdioServerParams(command="npx.cmd", args=["-y","@modelcontextprotocol/server-github"],env={"GITHUB_PERSONAL_ACCESS_TOKEN":os.getenv("GITHUB_PERSONAL_ACCESS_TOKEN")})
    tools = await mcp_server_tools(github_mcp_server)

    # Create an agent that can use the fetch tool.
    model_client = OpenAIChatCompletionClient(model="gpt-4o",api_key=os.getenv("OPENAI_API_KEY"))
    agent = AssistantAgent(name="github_assistant", model_client=model_client, tools=tools, reflect_on_tool_use=True, model_client_stream=True)  # type: ignore
    # Let the agent fetch the content of a URL and summarize it.
    await Console (agent.run_stream(task="Is there a repository named Autogen",cancellation_token=CancellationToken()))
    # print(result.messages[-1].content)

asyncio.run(main())

The output is :

---------- user ---------- Is there a repository named Autogen ---------- github_assistant ---------- [FunctionCall(id='call_1JsOJoUPn04qqIekNJfAhgtI', arguments='{"query":"Autogen"}', name='search_repositories')] GitHub MCP Server running on stdio ---------- github_assistant ---------- [FunctionExecutionResult(content='Error: unhandled errors in a TaskGroup (1 sub-exception)', name='search_repositories', call_id='call_1JsOJoUPn04qqIekNJfAhgtI', is_error=True)] ---------- github_assistant ---------- I encountered an issue while trying to search for a repository named "Autogen." However, you can manually search for it on GitHub using their search functionality. Just go to [GitHub](https://github.com) and enter "Autogen" in the search bar to see if such a repository exists. If you need further assistance or details, feel free to ask!

The same error is happening for the Github MCP server

[ekzhu]

Thanks. Looking into it.

[ekzhu]

Fix: #6125

[Omarchen-Good]

Excuse me, I met the same error,
My OS is windows 11, and in vscode + cline, I need to use the absolute path.
This MCP filesystem server is worked in vscode + cline, I excepted that Autogen works well too, but when I use Autogen to implement, it got the error as

---------- file_manager ----------
[FunctionExecutionResult(content='Error: unhandled errors in a TaskGroup (1 sub-exception)', name='write_file', call_id='call_heHvm1EmLQUKrqRS3I9Dyf9H', is_error=True)]
---------- file_manager ----------
Error: unhandled errors in a TaskGroup (1 sub-exception)

The full code is below:

import asyncio
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.models.ollama import OllamaChatCompletionClient
from autogen_ext.tools.mcp import StdioServerParams, mcp_server_tools
from autogen_agentchat.agents import AssistantAgent
from autogen_core import CancellationToken
from autogen_agentchat.messages import TextMessage
from autogen_agentchat.ui import Console


async def main() -> None:
    # Setup server params for local filesystem access
    server_params = StdioServerParams(
        command="D:\\Program Files\\nodejs\\node.exe",
        args=[
            "D:\\Program Files\\nodejs\\node_global\\node_modules\\@modelcontextprotocol\\server-filesystem\\dist\\index.js",
            "E:\\Work\\AI\\Cline"]
    )

    # Get all available tools from the server
    tools = await mcp_server_tools(server_params)

    # Create an agent that can use all the tools
    model_client_OpenAI = \
        OpenAIChatCompletionClient(model="gpt-3.5-turbo",base_url="XXXXXX",api_key="XXXXXX")
    
    agent = AssistantAgent(
        name="file_manager",
        model_client=model_client_OpenAI,
        tools=tools,  # type: ignore
    )

    # The agent can now use any of the filesystem tools
    stream = agent.on_messages_stream(
        [TextMessage(content="Create a file named 'test.txt'", source="user")],
        CancellationToken())
    await Console(stream)


if __name__ == "__main__":
    asyncio.run(main())

[appuk45]

I have updated the Autogen packages using pip (pip install -U "autogen-agentchat"). But the issue is still there.
`import asyncio

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import StdioServerParams, mcp_server_tools
from autogen_agentchat.ui import Console
import os
from autogen_core import CancellationToken
from dotenv import load_dotenv
load_dotenv()

async def main() -> None:
# Get the fetch tool from mcp-server-fetch.
github_mcp_server = StdioServerParams(command="npx.cmd", args=["-y","@modelcontextprotocol/server-github"],env={"GITHUB_PERSONAL_ACCESS_TOKEN":os.getenv("GITHUB_PERSONAL_ACCESS_TOKEN")})
tools = await mcp_server_tools(github_mcp_server)

# Create an agent that can use the fetch tool.
model_client = OpenAIChatCompletionClient(model="gpt-4o",api_key=os.getenv("OPENAI_API_KEY"))
agent = AssistantAgent(name="github_assistant", model_client=model_client, tools=tools, reflect_on_tool_use=True, model_client_stream=True)  # type: ignore
# Let the agent fetch the content of a URL and summarize it.
await Console (agent.run_stream(task="Is there a repository named Autogen",cancellation_token=CancellationToken()))
# print(result.messages[-1].content)

asyncio.run(main())`

`---------- user ----------
Is there a repository named Autogen
---------- github_assistant ----------
[FunctionCall(id='call_QkQPPn2vud7EyMVenIYdkOTL', arguments='{"query":"Autogen","perPage":1}', name='search_repositories')]
---------- github_assistant ----------
[FunctionExecutionResult(content='Error: unhandled errors in a TaskGroup (1 sub-exception)', name='search_repositories', call_id='call_QkQPPn2vud7EyMVenIYdkOTL', is_error=True)]
---------- github_assistant ----------
It seems there was an error when attempting to search for the repository. However, I can attempt a different approach or provide information based on available data.

From what I know up to October 2023, there is indeed a repository named "Autogen," specifically one notable project is under Microsoft called microsoft/Autogen. It is a project that involves using AI for generating code or other automation purposes. However, for the most updated or specific versions and other repositories with similar names, you might want to directly search on platforms like GitHub.

If you need further assistance or more detailed information about a specific Autogen repository, please let me know!`